Semantically-Aware Image Extrapolation

3. The computer-implemented method of claim 1, further comprising computing an image loss for the panoptic label map based on a center loss associated with the center coordinate and an offset loss associated with the offset for at least some of the plurality of output instances to identify partial instances among the plurality of output instances.

4. The computer-implemented method of claim 1, further comprising computing a discrepancy between the extrapolated semantic label map and a ground truth instance boundary map to refine the extrapolated semantic label map.

6. The computer-implemented method of claim 5, wherein refining the extrapolated semantic label map includes summing the local focal loss at the plurality of locations.

7. The computer-implemented method of claim 1, wherein the at least one mean characteristic comprises an average color.

9. The system of claim 8, further comprising a patch co-occurrence discriminator module configured to discriminate a texture from the input image and apply the texture to the outpainted region.

10. The system of claim 8, wherein the generator network module is further configured to compute an image loss for the panoptic label map based on a center loss associated with the center coordinate and an offset loss associated with the offset for at least some of the plurality of output instances to identify partial instances among the plurality of output instances.

11. The system of claim 8, wherein the POG network module is further configured to refine the extrapolated semantic label map based on a discrepancy between the extrapolated semantic label map and a ground truth instance boundary map.

12. The system of claim 11, wherein the POG network module is further configured to refine the extrapolated semantic label map based on a local focal loss between the extrapolated semantic label map and the ground truth instance boundary map.

13. The system of claim 12, wherein the POG network module is further configured to refine the extrapolated semantic label map based on a sum of the local focal loss at a plurality of locations.

14. The system of claim 8, wherein the at least one mean characteristic comprises an average color.

17. The non-transitory computer-readable medium of claim 15, wherein the program code is further executable to perform an operation of computing an image loss for the panoptic label map based on a center loss associated with the center coordinate and an offset loss associated with the offset for at least some of the plurality of output instances to identify partial instances among the plurality of output instances.

18. The non-transitory computer-readable medium of claim 15, wherein the program code is further executable to perform an operation of computing a discrepancy between the extrapolated semantic label map and a ground truth instance boundary map to refine the extrapolated semantic label map.

20. The non-transitory computer-readable medium of claim 19, wherein the operation of refining the extrapolated semantic label map includes an operation of summing the local focal loss at the plurality of locations.